Investigating the Environmental Kuznets Curve

hypothesis using Environmental Performance Indices

Master Thesis

Department of Economics

Lund University

Spring 2011

Author: Olena Kashyna

Supervisor: Susanna Thede

2

Abstract

This essay investigates the Environmental Kuznets Curve (EKC) hypothesis using the

Environmental Performance Index (EPI). The index incorporates various environmental

indicators related to human health as well as the human impact on ecosystem vitality.

Therefore it assesses environmental quality better than separate environmental indicators,

which are usually used to test the EKC hypothesis. We run OLS regression on cross-section

data including developed and developing countries for the years 2006 and 2008. Our

estimates confirm the EKC hypothesis: We find that the environmental quality initially

worsens but eventually improves with an increasing level of per capita income. The impacts

of income inequality and the level of political freedom on the environmental performance are

also studied. The overall results suggest that there is a strong positive impact of the level of

political freedom on the environmental performance, while income inequality appears to have

an ambiguous impact.

Keywords: Environmental Kuznets Curve, Environmental Performance index, Income

inequality

Course: NEKM02, Master Thesis in Economics, 15 ETCS

3

Contents

1. Introduction…………………………………………………………………………… 4

2. Theoretical framework………………………………………………………………… 6

2.1 The Kuznets curve……………………………………………...……………. 6

2.2 The Environmental Kuznets curve………………………………….………... 6

2.3 The use of environmental indices in EKC estimations…..…………………… 9

2.4 Empirical evidence on the EKC hypothesis…………………………………. 10

2.5 Income inequality and the level of political freedom as additional factors

influencing environmental quality ……..…………………………………… 11

2.5.1 Income inequality…………………….…………………….………..11

2.5.2 Political rights and democracy………….…………………….…….. 13

3. Methodology………………………………………………………………………….... 15

4. Data……………………………………………………………………………………. 17

4.1 The Environmental Performance Index……………………………………… 17

4.2 Data on Per capita income, Political freedom and Income inequality...……… 18

4.3 Descriptive statistics………………………………………….………………. 19

5. Empirical results and analysis…………………………………………………………. 21

5.1 Estimations of the EKC hypothesis for the year 2006……………………..... 21

5.2 Estimations of the EKC hypothesis for the year 2008……………………….. 26

6. Conclusions……………………………………………………………………………. 29

References………………………………………………………………………………... 31

Appendix………………………………………………………………………………..... 34

4

1. Introduction

The promotion of economic growth has been a crucial part of development planning and

policymaking in the second half of 20th century. In the last decades, this policy focus has

been complemented by increased attention to sustainability and welfare issues, as a response

to increasing recognition of the problems which modern nations, especially developing ones,

encounter. These problems include, among others, income inequality, environmental decay

and resource depletion.

In his classic paper, Kuznets (1955) presented the idea of an inverted U-shape

relationship between income inequality and economic development. According to his

hypothesis, at low levels of economic development, economic growth causes increased

inequality, but after some turning point further economic development decreases income

inequality. Recently the concept of the environmental Kuznets curve (EKC) has been studied

in the scientific literature, investigating the inverted U-shape relationship between economic

development and the level of environmental quality. According to the EKC hypothesis,

environmental quality initially worsens but eventually improves with increasing level of

economic development.

The amount of scientific articles dedicated to the EKC hypothesis is impressive: For

example, one of the pioneering studies by Grossman and Krueger (1995) has been quoted

more than 2150 times.1 However, empirical estimations on the subject provide ambiguous

results. To test the EKC hypothesis, most researchers use different proxies for environmental

degradation, such as data on different pollutant emissions (sulphur or carbon dioxide), as well

as organic water pollution, deforestation rates, percentage of land allotted to protected areas

etc. A singular indicator can explain only some aspect of environment quality, but is unable to

address the overall state of environment. Until recently, no composite indicator of

environmental quality existed, which can be explained by the scarcity of the environmental

data. Only a few studies tried to construct such an index: Färe et al (2004), Jha and Bhanu

Murthy (2003), Xiaoyu et al (2011), but their approach is still fragmentary and does not

access the environmental quality as a whole.

The recently launched Environmental Performance Index (EPI) offers a new

comprehensive approach towards the measurement of environmental performance. This index,

developed by Yale University and Colombia University in collaboration with the World

Economic Forum and the Joint Research Centre of the European Commission, quantifies and

1 According to Google Academy http://scholar.google.se

5

provides a benchmark for countries‟ environmental policies. As of May 2011 three EPI

reports have been released, based on 2006, 2008 and 2010 data.

The aim of this essay is to assess the EKC hypothesis using Environmental Performance

Indices. To our knowledge, there has been only one similar study deploying EPI in the EKC

estimations (Yoshioka, 2010). While their study is based on the 2006 EPI, we perform our

estimations on the cross-section data for the two EPI indices: the Pilot 2006 EPI, and the 2008

EPI. Moreover, we use improved methodology and account for other factors which may affect

environmental performance apart from the income level.

Even though the EKC may hold, it doesn‟t automatically follow that economic growth is

by itself a remedy to the environmental problems or that there is no space for policy

interventions apart from supporting economic growth. Therefore, we test some additional

hypotheses, developed in the EKC literature – namely, whether income inequality and the

level of political freedom influence the shape of the EKC. It will allow us to provide some

relevant policy recommendations: for example, whether inequality reduction policies can have

a positive spillover effect on environmental quality or whether democratization may promote

it.

The study will contribute to the existing literature on the relationship between per capita

income and environmental quality, and will try to provide relevant policy advice. Therefore, it

will be of interest for researchers, as well as practitioners and policy-makers.

The rest of the paper is organized as follows: Section 2 introduces the theoretical

framework, section 3 discusses our methodology, the data used in our empirical investigation

is presented in section 4. Section 5 presents and discusses the empirical results, and section 6

provides the main conclusions and policy implications of this study.

6

2. Theoretical Framework

This section provides the theoretical framework for our empirical analysis. It presents the

conceptual background of the Environmental Kuznets Curve, theoretical explanations of its

shape and additional factors which may affect it. We also summarize the existing empirical

evidence on the subject.

2.1 The Kuznets Curve

In his classic paper, Kuznets (1955) presented the idea of the inverted U-shape

relationship between income inequality and economic development, as measured by per

capita income. According to his hypothesis, at low levels of economic development,

economic growth causes increased inequality, but after some turning point with further

economic development the distribution of income becomes more equal. Kuznets discussed the

mechanisms underlying such a relationship and introduced empirical evidence based on time-

series data for the UK, the US and Germany. The Kuznets hypothesis stimulated a large

number of theoretical and empirical studies, which provided more solid theoretical grounds

and empirical evidence on the existence of the Kuznets curve.2

2.2 The Environmental Kuznets Curve

In the 1990s, the Kuznets curve was reconsidered in a different context. Some empirical

evidence showed that the level of environmental degradation and per capita income follow an

inverted U-shaped pattern, just as income inequality and per capita income in the original

Kuznets Curve. Therefore this pattern was named the Environmental Kuznets Curve (EKC).

The first empirical EKC studies comprised three independent working papers by Grossman

and Krueger (1993), Shafik and Bandyopadhyay (1992) and Panayotou (1993), all of which

reached the same conclusion, namely, that relationship between some pollution indicators and

per capita income can be described by an inverted-U curve.3 Panayotou (1993) was the first

one to use the „Environmental Kuznets Curve‟ term and since then the EKC has become a

standard tool to describe the relationship between the level of environmental quality and the

level of per capita income. Figure 1a illustrates the original Kuznets Curve and Figure 1b

presents the EKC. There are several arguments to explain the shape of EKC in the existing

literature, which we review below.

2 Kijima et al (2010), pp.1188–1189.

3 Dinda (2004), p. 433

7

2.2.1 Income elasticity of the demand for environmental quality.

As people‟s income and standard of living increases, their priorities change and they care

more for the quality of environment. Demand for a healthier and cleaner environment leads to

structural changes in the economy and reduces environmental degradation: More is spent on

cleaner technologies, people choose products which are less harmful for the environment,

donate more to environmental organizations and create political pressure for environmental

protection and regulations. This mechanism is called the “pure income effect” or the

“abatement effect” in the literature.4 An additional argument, presented by Dasgupta and

Laplante (2002), states that as per capita income increases, the marginal propensity to

consume should decline or at least be constant, which also helps explaining the shape of EKC.

2.2.2 Scale, technology and composition effects

On the production side, economic growth affects the environment through scale,

technology and composition effects (Grossman and Krueger, 1991). Scale effect implies that

more resources are used as the output increases, which leads to more waste and harmful

emissions. Through this effect economic growth has a negative impact on the environmental

quality. The composition effect influences the state of environment in an opposite way: As

income increases, the production structure changes towards less environmentally harmful

activities. At earlier stages of economic development, environmental degradation tends to

increase when the production structure changes from agricultural to industrial. However, it

tends to fall as economy moves towards more services and knowledge-based technology-

intensive industries. The technology effect of economic growth is essentially the same: As a

nation becomes more affluent, more resources are spent on R&D and dirty technologies are

replaced by cleaner ones, which improves the environmental quality. Figures 2a-c summarize

and illustrate these different effects.

4 Panayotou, (2003), p.52

Per capita income

Income

inequality

Environmental

degradation

(pollution)

Figure 1a Original Kuznets Curve Figure 1b Environmental Kuznets Curve

Per capita income

8

The EKC suggests that the scale effect dominates at the initial stages of an economy‟s

growth, but after some turning point the positive composition and technology effects, together

with the income effect and the demand for pollution abatement prevail, and hence

environmental quality improves (Vukina et al., 1999).

2.2.3 International trade

International trade can have offsetting environmental effects. On one hand, its impact can

be negative through the scale effect: Trade often expands production possibilities of an

economy, in which case the environmental degradation intensifies. On another hand, trade can

have a positive impact on the environmental quality through the technology effect: As income

rises with trade, more resources are directed to development of cleaner technologies. The

impact of the composition effect is ambiguous: Though international trade pollution from the

production of pollution-intensive goods decreases in some countries, it increases in others.

The composition effect is further related to the hypotheses of displacement and pollution

havens. Under the displacement hypothesis, EKC records a displacement of environmentally

harmful industries to less developed economies. Therefore, the EKC reflects not the change in

consumption patterns of certain nations, but the change in international specialization, where

poorer countries specialize in more energy- and other resource-consuming activities, while

developed countries specialize in services and production with cleaner technologies. The

pollution haven hypothesis is quite similar and refers to the situation when low environmental

standards become a comparative advantage and stimulate multinational firms to shift their

environmentally harmful production to poorer countries where environmental regulation is

less strict.5

5 Dinda (2004), pp. 435-437

Per capita income Per capita income

En

vir

on

men

tal

deg

rad

atio

n

Per capita income E

nv

iron

men

tal

deg

rad

atio

n

En

vir

on

men

tal

deg

rad

atio

n

Figure 2a Figure 2b Figure 2c

The scale effect The composition and The income effect

technology effect

9

2.3 The use of environmental indices in the EKC estimations

There have been few studies in the EKC literature which focused not on the separate

environmental indicators, but tried to assess the overall environmental performance.

Some studies have been trying to assess the production process and develop the index of

environmental efficiency. Zaim and Taskin (2000) derive an environmental efficiency index

based on a production approach that differentiates between environmentally desirable and

undesirable outputs. The authors construct a frontier which represents the best-practice

technology using data on inputs (aggregate labour and total capital stock), desirable output

(real GDP) and undesirable output (represented by CO2 emissions). Based on this approach,

only the countries that use the most efficient technologies are located on the production

frontier, while others are behind it. The authors‟ evidence supports the EKC: They conclude

that environmental efficiency deteriorates with lower levels of income, but then improves

after a turning point. The main limitation of the study is that it focuses only on OECD

countries.

A similar methodology to construct an environmental performance index is adopted by

Färe et al (2004). However, undesirable output in their estimations is represented by several

pollutants: carbon dioxide, nitrogen and sulphur oxide emissions. Using data on OECD

countries, they find no evidence of EKC. The implications of their study are very important:

While the EKC may be found for separate pollutants as shown in a study by Zaim and Taskin

(2000), there may be no clear-cut relationship between the overall environmental degradation

and per capita income if one accounts for several pollutants simultaneously.

Jha and Bhanu Murthy (2003) develop a composite environmental degradation index

using principal component analysis and data on six environmental variables (such as CO2

emissions per capita, deforestation rate, annual per capita fresh water withdrawals etc.). While

estimating the EKC hypothesis, they link the index not to the level of income, but to the

Human Development index (HDI), which in the authors‟ view better reflects the level of

economic development. Their empirical estimates show that the environmental degradation

with respect to economic development doesn‟t follow an inverted U-path, but has an inverted

N-shape, which means that environmental degradation first decreases with income, then

slightly increases to start decreasing again at the highest levels of economic development.6

Moreover, the authors conclude that there are large inequalities in the contribution to the

6 Jha and Bhanu Murthy (2003), p.365

10

global environmental degradation across countries, with the small number of developed

countries accounting for more than 50% of it.

Xiaoyu et al (2011) construct an environmental pollution index, taking into account 24

indicators within six dimensions of pollution: industrial wastewater, industrial waste gas,

industrial solid waste, domestic wastes, air quality and energy consumption. Using panel data

on thirty Chinese provinces, the authors test the EKC hypothesis and find empirical support

for it.

To our knowledge, Yoshioka (2010) is the only prior study using EPI in EKC estimations.

He deploys the cross-section data for the 2006 EPI. The author doesn‟t confirm the EKC for

the overall index, but only for several indicators within the index. Such results, however, may

be driven by misspecification of the model and insufficient data transformation: For example,

the author doesn‟t use the logarithmic values of variables in order to smooth the variables‟

distribution, which can substantially influence the estimation results.

2.4 Empirical evidence on the EKC hypothesis

Dinda (2004), He (2007) and Kijima (2010) review the existing EKC literature and

provide critical surveys on the theoretical explanations and empirical evidence supporting the

EKC hypothesis. They all conclude that the results of estimations are quite controversial and

depend heavily on the choice of variables, sample and methodology. The EKC relationship

typically holds for energy-related air pollutants, such as sulfur dioxide and particulate matter,

which may be due to the fact that such pollutants are subject to more regulation. For other

environmental indicators, such as water pollution, municipal waste, CO2 or energy use,

evidence of an EKC is quite inconsistent: These indicators either increase monotonically with

per capita income or have turning points at very high levels of per capita income. Additionally,

indicators that have a direct impact on human health (for example, access to drinking water

and urban sanitation) tend to improve with increasing per capita income.7 Some empirical

findings suggest that certain pollutants follow an N-shaped relationship with income, which

means that environmental degradation follows the inverted U-path initially, but beyond a

certain level of income the relationship between the environmental degradation and income

becomes positive again. This case is illustrated in Figure 3.1.

7 Dinda (2004), p. 441-442

11

Moreover, it may be that empirical evidence supporting the EKC hypothesis is applicable

only for developed nations, since the studies that identify EKC relationship mostly are

focused on OECD countries. For those studies that consider time-series or panel data on the

developing nations, the empirical evidence is ambiguous, and the turning points of the EKC

often don‟t exist or are found at substantially higher levels of per capita income than for

developed nations.8 Additionally, the EKC for developed nations may be confirmed because

of the composition effect, which forces polluting industries out of the affluent countries to the

poorer ones.

Besides the per capita income, other factors can also influence environmental change,

such as the population density, the income inequality, the production structure of the economy,

historical events, the degree of political freedom and democracy etc. If they do, this would

imply that the environmental quality does not improve automatically as per capita income

grows.9 We will now analyze in more detail the factors that have been argued to have the

strongest impact on the environmental quality apart from the level of income in the

predominant part of these studies, namely income inequality and the level of political freedom.

2.5 Income inequality and the level of political freedom as additional factors

influencing environmental quality

2.5.1 Income inequality

Combining the implications of the original Kuznets curve and the EKC, the decrease in

income inequality could be expected to improve the state of environment. Boyce (1994)

establishes a theoretical framework for the analysis of the relationship between income

inequality and environmental quality. He argues that the scope of environmentally harmful

economic activities depends on the balance of power between those who derive net benefits

8 He (2007), p.10-11

9 Gallagher and Thacker (2008)

Per capita income E

nv

iron

men

tal

deg

rad

atio

n

Figure 3.1 Cubic relationship between per capita income

and the environmental degradation

12

from such activities and those who bear the net costs; greater inequalities and imbalances in

power and wealth lead to more environmental degradation. In contrast to this political

economy argument, Heerink et al (2001) argue that if an inverted U-shaped EKC holds on the

household level, then lower income inequality increases the level of environmental

degradation. This happens because richer households are located on the downward sloping

part of the Environmental Kuznets curve and poorer households are located on the upward

sloping part of the EKC, leaving more households closer to the peak of the environmental

degradation with higher income equity. Magnani (2000) argues that the downward sloping

segment of the Environmental Kuznets Curve emerges not only conditioned by the country‟s

ability to pay for environmental quality, but also its willingness to do so. The willingness to

pay for environmental quality depends on the relative income, or on how much a

representative individual‟s income differs from the average income. Thus, greater income

inequality negatively affects environmental policy decisions. Marsiliani and Renström (2002)

derive conditions on individual preferences and technology that give rise to a negative

correlation between income inequality and environmental quality. They present a model in

which individuals have different levels of income, and where a representative, elected by the

majority, decides on the level of pollution and redistribution taxes. The authors show that if

income of the decisive individual is lower than the average income, then a higher

redistributive tax and a lower pollution tax will be preferred.

Kempf and Rossignol (2007), Borissov et al (2010) use similar reasoning based on the

median voter framework, and show that inequality can be harmful for the environment. They

argue that there is a trade-off between the pollution-generating economic growth and

environmental quality. They show that the poorer the median voter is relatively to the average

individual, the fewer resources will be devoted to the environment, and economic growth will

be preferred.

Besides the theoretical contributions mentioned above, there are a number of empirical

studies that consider income inequality in EKC estimations. Hill and Magnani (2002)

examine the conceptual and empirical basis of the EKC and argue that in estimating a simple

EKC relationship, the omitted variables problem arises, and empirically show that income

inequality is a key variable that needs to be included in EKC estimations.

Bimonte (2002) empirically tests the hypothesis of the EKC existence for the percentage

of protected areas within the national territory, and stresses that the income distribution,

among other variables such as education and information accessibility, may play a

fundamental role in determining environmental quality.

13

Holland et al. (2009) use quite an unusual measure of environmental deterioration,

investigating whether income inequality helps explaining biodiversity loss, and find that it

indeed has a strong negative effect on the proportion of threatened species.

Clement and Meunie (2010) examine the relationship between social inequality and

pollution, where social inequality incorporates both income inequality, as measured by the

Gini coefficient, and political power inequality, as measured by the Freedom House political

rights index. The authors provide a survey of theoretical approaches on the research topic, and

empirically estimate the inequality impact on pollution, using panel data on 83 developing

and transition countries. They confirm the EKC hypothesis for SO2 emissions, but for water

pollution the relationship with income follows an N-shaped curve. According to their findings,

higher income inequality explains higher water pollution levels in developing countries, and

higher levels of political freedom is associated with lower levels of pollution.

2.5.2 Political rights and democracy

Even though some empirical findings conclude that the relationship between

environmental degradation and income has an inverted U-shape, it doesn‟t automatically

follow that economic growth by itself can be regarded as a remedy to environmental problems

or that there is no space for policy interventions apart from supporting continuous economic

growth. On the contrary, as Grossman and Krueger (1995) argue, “the strongest link between

income and pollution in fact is via an induced policy response. As nations or regions

experience greater prosperity, they citizens demand more attention to be paid to the

noneconomic aspects of their living conditions.”10

The effect of political rights or democracy on the environmental quality is ambiguous. Li

and Reuveny (2006) present the survey of the contradicting theoretical arguments, as well as

empirical evidence on this issue. Democracy can improve environmental quality by raising

public awareness and encouraging environmental legislation, because environmental interest

groups are more successful at informing people and organizing them in a democracy than in

an autocracy (autocratic regime can censor information flows). Moreover, democracies are

better at representing environmental needs of the citizens through environmental political

parties and groups, which can influence public policy. Gallagher and Thacker (2008) argue

that democratic governments are more accountable for their actions and tend to cooperate

more among themselves and sign international treaties to protect the environment.

10

Grossman and Krueger (1995), p. 372

14

Democratic regimes can also have a negative impact on the state of environment through

a number of other mechanisms. One of them is the so-called “Tragedy of Commons”, when

free individuals and groups can overexploit common resources, ignoring the environmental

damage of their economic actions. Moreover, democracies tend to be market economies,

where business groups can have much more political power than environmentalists, because

democratic leaders are accountable to business groups that support their coming to power.

Since the interests of such groups prioritize profit-maximization to environmental concerns,

this bears potential hazards for the quality of environment.11

Boyce and Torras (1998) include a measure of political rights and civil liberties in

estimations of the EKC and find that such rights have positive, statistically significant impact

on the environmental quality, as measured by different pollution variables. An especially

strong effect is found for low-income countries, while for high-income countries the effect is

weaker. The authors explain this finding by the implications of Kuznets‟ original hypothesis:

At lower levels of income, there is also more power inequality. Under high levels of power

inequality those who benefit from pollution-generating activities tend to be more powerful

than those who bear the costs, which results in higher pollution levels.

Li and Reuveny (2006) report that the existing empirical evidence on the EKC is mixed

and present their own estimation. It is based on a broader sample size covering five main

types of human-induced environmental degradation: carbon dioxide emissions, nitrogen oxide

emissions, land degradation, deforestation, and organic pollution in water, as well as two

composite environmental indicators. The authors find that a stronger democracy reduces

environmental degradation, although the size of this effect varies with different environmental

indicators.

Gallagher and Thacker (2008) treat democracy as a cumulative phenomenon and find a

strong positive relationship between the long-term democracy (existing for a long time) and

environmental quality. Buitenzorgy and Mol (2011) investigate the impact of democracy on

deforestation rates and find evidence that an inverted U-shaped relationship exists between

these variables. Their empirical findings suggest that countries in democratic transition

experience much higher deforestation rates than non-democracies or mature democracies.

Moreover, the authors show that democracy has a stronger explanatory power than per capita

income in explaining deforestation rates.

11

Li and Reuveny (2006), p.937-939

15

3. Methodology

This section provides a description of the model we are estimating. We describe the

standard model estimated in the EKC literature, and provide a motivation for our alternative

modifications of the model.

A reduced-form equation is typically used in the EKC estimations, where the level of

environmental degradation or pollution is regressed upon the per capita income, the squared

value of per capita income and additional determinants. This is done instead of modeling

structural equations where environmental regulations, technology and industrial composition

are related to GDP, and pollution is related to the regulations, technology and industrial

composition. As Grossman and Krueger (1995) argue, the main advantage of the reduced-

form approach is that it provides the net effect of per capita income on pollution, which in the

case of structural approach would have to be estimated stage by stage and therefore would be

more dependent upon the precision and bias at every stage. Moreover, under a reduced-form

approach there‟s no need in collecting the data on pollution regulations and technology, which

is important given the scarcity of such data.12

The following reduced form model is typically used to test EKC hypothesis:13

where the i and t subscripts denote the country and time, y is a certain environmental

degradation indicator, α is a country-specific effect, is time-specific effect, x is per capita

income and Z is a vector of additional explanatory variables which can influence

environmental quality. Often logarithmic values of environmental indicators and income are

used in the estimations. This is done in order to smooth the distribution of the data, and

because the predicted y-variable only can take on positive values (some environmental

damage is always expected to occur).14

The EKC hypothesis is confirmed if β1>0, β2<0 and β3=0, where β1>0 captures the linear

increase of environmental degradation with income, and β2<0 indicates the existence of the

function‟s maximum, or “turning point”. If β1>0, β2<0 and β3>0, the relationship between the

variables is cubic or N-shaped.

12

Grossman and Krueger (1995), p.359 13

Dinda (2004), p.440 14

Stern (2003), p.3

16

The turning point of the EKC is estimated as .15

In our study due to data limitations we can‟t use panel or time-series data, therefore we

drop t-subscript from the equation provided above. We use a simple OLS regression on cross-

section data that incorporates the following basic model:

(1)

where is the EPI and GDP/P is GDP per capita.

In order to account for heteroskedasticity, we use White‟s heteroscedasticity consistent

covariance estimator. In the presence of heteroskedasticity of unknown form, which it likely

to be the case for cross-sectional data, it provides consistent estimates of the coefficient

covariances.16

Since the EPI is a measurement of environmental quality, and not of environmental

degradation, the expected coefficients are reversed if the EKC hypothesis is valid: β1<0, β2>0

and β3=0. In other words, environmental performance is expected first to decrease with an

increasing level of per capita income, but then to improve after turning point.

Since we want to test for the impact of income inequality as well as the level of political

freedom, we estimate different specifications of the basic models with additional explanatory

variables:

(2)

where is a vector of additional explanatory variables. We estimate regressions with

each of the additional variables added separately, but also examine the joint effects and the

interaction effects of the income inequality and the level of political freedoms. Moreover, we

examine the difference in the impact of variables for different country groups in the sample.

15

To find the turning point, we differentiate the estimated equation with respect to per capita income and set

it to be equal to zero: ; ; 16

Verbeek (2004), pp. 86-88

17

4. Data

In this section we introduce our data, and present our variable construction. We also

present descriptive statistics.

4.1 The Environmental Performance Index

The Environmental Performance Index is constructed in collaboration between Yale

University (Center for Environmental Law and Policy) and Columbia University (Center for

International Earth Science Information Network), with support from the World Economic

Forum and the Joint Research Centre of the European Commission. The index is based on

proximity-to-target methodology, which is focused on a set of environmental outcomes linked

to policy goals. By formulating specific targets and measuring how close each country comes

to them, the EPI provides a basis for policy analysis and for evaluating environmental

performance, and also facilitates cross-country comparisons.

The index contains two main categories: environmental health and ecosystem vitality.

Each of the categories in turn consists of a number of subcategories, such as air quality, water

resources, biodiversity and habitat, natural resources, climate and energy. Specific indicators

within subcategories are the basic elements of the EPI. Each indicator, subcategory and

category is assigned a certain weight in an overall index. Appropriate weights for each

indicator are identified on the basis of the principle component analysis, with refinements and

modifications suggested by the expert group of the EPI team.17

Relevant long-term public health or ecosystem sustainability goals are identified for each

indicator. These targets are set as benchmarks, common for all the countries, and are based on

international agreements, standards established by international organizations and authorities,

or prevailing expert judgment.18

To make indicators comparable, each is converted to a

proximity-to-target measure with a range from 0 to 100. Extreme indicator values are

winsorized to avoid skewed aggregations.19

Environmental Performance Index for years 2006 and 2008 is obtained from the index

website, launched by the Yale Center for Environmental Law & Policy. The Pilot EPI 2006

uses 16 indicators and covers 133 countries. The 2008 EPI has similar structure to the 2006

EPI version, but deploys 25 indicators: some of the 2006 EPI indicators are removed, and

some are added. EPI 2008 is estimated for 149 countries. The 2010 EPI ranks 163 countries

17

Esty et al. (2008), pp. 22-23 18

Esty et al (2006), p. 9 19

Winsorization is a statistical technique, which adjusts extreme values in the sample by setting them to a specified percentile of the data.

18

on 25 indicators (also slightly modified compared to the 2008 version), but we don‟t use it in

our estimations, since the data on additional variables is not available for year 2010 yet.20

Appendix A provides the composition of the 2006 and 2008 EPI. The methodology

applied in 2008 has been substantially improved since the 2006 Pilot version, and since these

indices aren‟t directly comparable, we need to perform individual estimations for the two

years.

Where applicable, the EPI targets are population-based. For example, for the “Adequate

sanitation” indicator, the target is set by "100% of population having access to adequate

sanitation", and for the “Greenhouse gas emissions” the target is expressed in per capita

values; but for such an indicator as the “Water stress”, the target is set by "0% territory under

water stress" and is rather territory- than population-based. Since the population size is

already incorporated in the EPI structure, there is no need to account for the size of the

population in our estimations.

As the index creators concede themselves, the main limitation of the EPI is the lack of

time-series data and the inability to track change in environmental performance over time.

However, it still allows us to perform cross-country analysis to explain the overall

environmental performance.

4.2 Data on Per capita income, Political freedom and Income inequality

Original datasets on the EPI also include data on GDP per capita, but their data is not

contemporaneous: the 2006 EPI and the 2008 EPI include data on 2005 GDP per capita. In

our estimations we use data on income for years 2006 and 2008, taken from the World

Development Indicators database: GDP per capita based on purchasing power parity (PPP),

evaluated in constant 2005 international dollars. This way, our estimations capture long-run

level effects in a way consistent with the theoretical EKC relationship.

As a measure of income inequality we use the Gini index obtained from the World Bank

database. The Gini coefficient is a standard measure for the inequality of a distribution, and is

commonly used to measure the inequality of income among individuals or households in an

economy. It ranges from 0 to 1 (or from 0 to 100%), where zero value represents perfect

equality, while 1 implies perfect inequality.21

Since the available data on the Gini coefficient

is very fragmentary, we use linear interpolation method for the period 1995-2009, as well as

Eurostat and OECD statistical databases to fill in gaps in the data.

20

Only data on the Freedom House index is available for year 2010 as to date. 21

World Development Indicators: http://data.worldbank.org/

19

We deploy two different measures of the level of democracy. One of them is the Polity

IV measure, obtained from The Center for Systemic Peace website. Polity IV is an indicator

of the level of democracy, which is computed by subtracting 10-point AUTOC index

measuring autocratic characteristics from the 10-point DEMOC index measuring democratic

characteristics. Thus a composite indicator ranging between -10 and 10 is obtained. For those

countries which are assigned “standardized authority scores”: -66 standing for foreign

interruption, -77 standing for interregnum or anarchy, and -88 standing for “transitions”, the

polity scores are treated in accordance with index manual. We follow Gurr et al (2010) and

transform these data points as follows: -66 and -88 are transformed to missing values, and -77

score is set to be equal to zero.22

This is done in order to avoid extreme outliers in the variable

distribution.

Li and Reuveny (2006) argue that while continuous measures of democracy such as

Polity IV are informative, the effect across a range of values along the scale may not be

constant. For example, the effect of democracy rising from -10 to -5 may not be the same as

the effect of it increasing from 0 to 5.23

Therefore, following the authors, we use dichotomous

measures of democracy and autocracy based on the Polity IV indicator. We create dummy

variables for democratic and autocratic regimes, defining a country as democratic if Polity IV

measure is above 5 and as autocratic if its score is below -5.24

Another measure used to assess the level of political freedom is Freedom House

indicators of the state of civil and political rights. Each of the two indicators (Civil rights and

Political rights) ranges from 1 to 7, where 1 indicates the most free and 7 indicates the least

free country. Since the original Freedom House index depicts that higher values indicate

lower levels of political freedom, but higher values correspond to higher levels of political

freedom with Polity IV index, we perform a simple transformation to make the indices easier

to compare and analyze. We take a simple average of the two Freedom House indicators,

subtract this average from 7, and add 1, obtaining a 1-7 scale, where higher values of index

reflect higher level of political freedom.

The two indices measure essentially the same thing: a simple correlation coefficient

between these measures is very high and equals to 0,90 in 2006 and 0,87 in 2008.

22

Gurr et al. (2010), p.17 23

Li and Reuveny (2006), pp.941-942 24

As Li and Reuveny (2006) point out, the thresholds are quite arbitrary in the sense that slightly higher or lower values of Polity IV index could also be chosen, but the values of 5 and -5 are the most common in scientific literature.

20

In our further estimations, test for the possibility that income inequality and the level of

political freedom effects may differ depending on the country‟s level of development.

Therefore, we subdivide our country sample into three income groups: low-, middle- and

high-income countries. We deploy the World Bank definitions and use the following

threshold values: less than 1000 dollars per capita for low-income countries; between 1000

and 12000 dollars per capita for middle-income countries; above 12000 for high-income

countries. Using this definition, our sample for the year 2006 contains 15 low-income, 74

middle-income, and 39 high-income countries. For the year 2008, there are 13 low-income,

85 middle-income and 46 high-income countries. The list of these countries is provided in

Appendix C. Appendix D contains the list of the countries for which the Gini coefficient is

available, divided according to their level of income.

4.3 Descriptive statistics

Tables B1 and B2 in Appendix B present descriptive statistics for the variables entering

our estimations. As already mentioned, the two indices are not directly comparable between

each other due to the differences in methodology and sample sizes. We can observe the

increase in the mean value for the EPI from 64,45 in 2006 to 71,87 in 2008, which may be

partially explained by the change in the index methodology, but may also reflect improvement

in the global environmental performance. Moreover, the 19 countries added to the 2008 EPI

on average have quite high EPI scores.25

We can also look at the separate indicators within EPI that remain unchanged within

index framework, to draw some conclusions on the dynamics of different aspects of

environmental performance. Thus, we can see that such indicators as Adequate water

sanitation, Access to drinking water, Indoor air pollution and Urban particulates improved

from 2006 to 2008, while the Agricultural subsidies deteriorates; the rest of the indicators are

not comparable through the two indices.

25 These include Belarus, Belize, Bosnia and Herzegovina, Botswana, Croatia, Dijibouti, Eritrea, Estonia, Fiji,

Guyana, Iraq, Kuwait, Latvia, Lithuania, Luxembourg, Macedonia, Mauritius, Solomon Islands and Uruguay (average EPI for these countries is 72,6); Gambia, Liberia and Suriname, present in the 2006 EPI, don’t have the EPI score for year 2008.

21

5. Empirical analysis and results

In this section we report the empirical results of the methodology applied to our data and

discuss them. We estimate different specifications of the basic model with additional

explanatory variables for two years: 2006 and 2008. These are the two years for which cross-

section data on all variables entering our model are available. Microsoft Excel was used to

transform the data, and all the estimations are performed in E-Views 7.0.

5.1 Estimations of the EKC hypothesis for the year 2006

Table 5.1.1 presents the results of estimations for the year 2006. Specification I is our

basic model as described by regression equation (1). Specification II includes the

environmental effects of income inequality, as captured by the Gini coefficient. Specification

III includes the Polity IV measure for the level of political freedom, while specification IV

uses the Freedom House index. In specification V and VI we use the two dummy variables

Autocracy and Democracy.

We confirm the EKC hypothesis for all model specifications, since β1<0 and β2>0, and

these parameter estimates are statistically significant. In other words, environmental

performance decreases with an increasing level of per capita income, but then to starts to

improve after the turning point. Moreover, there is evidence of an N-shaped EKC, since the

parameter estimate β3 is positive and statistically significant, although the impact of this

additional non-linear element is quite small. The model has high explanatory power, since it

explains about 66% of the variation in the dependent variable.

Both the Polity IV and Freedom House measures of the level of political freedom are

statistically significant and suggest that an increase in political freedom has a positive impact

on the overall environmental performance. The magnitude of these effects is similar for both

measures. Moreover, the dichotomous measures of autocracy and democracy are statistically

significant, and suggest that if a country has a democratic regime, it is more likely to have

better environmental performance, while if it is an autocracy, then it is more likely to have

worse environmental performance. The size of the effects, however, is small compared to the

income effects.

The model specification with the Gini coefficient, provides insignificant results for the

income measures and thus for the EKC hypothesis. The only statistically significant

coefficients in this specification are the intercept and the Gini coefficient. On one hand, this

may call for a different model specification. Heenrik et al (2001) find similar effects for

deforestation rates: when income inequality is included into the EKC model, income effects

22

are not significant, while income inequality is. The authors provide the following explanation

for this result: the EKC studies excluding income inequality may implicitly estimate the

original Kuznets curve, i.e. the impact of income on income inequality, and hence on the

environmental degradation.26

Following the authors, we also estimate the original Kuznets

curve for our data and find support for this proposition.27

Table 5.1.1 Estimations the EKC hypothesis with additional variables for the year 2006

Model specification Explanatory variable I II III IV V VI

Constant

11,52***

(4,24)

6,21*

(1,95)

10,42***

(4,08)

9,61***

(3,30)

11,07***

(4,06)

11,01***

(4,53)

ln GDP per capita -3,04***

(-3,11)

-1,15

(-0,98)

-2,65***

(-2,88)

-2,41**

(-2,32)

-2,87***

(-2,94)

-2,84***

(-3,23)

(ln GDP per capita)2

0,38***

(3,32)

0,16

(1,10)

0,34***

(3,11)

0,32**

(2,58)

0,36***

(3,16)

0,36***

(3,43)

(ln GDP per capita)3 -0,02***

(-3,38)

-0,01

(-1,08)

-0,01***

(-3,20)

-0,01***

(-2,71)

-0,01***

(-3,23)

-0,01***

(-3,47)

Gini coefficient - 0,004***

(2,75)

- - - -

Polity - - 0,007***

(4,13)

- - -

Freedom House - - - 0,02***

(3,15)

- -

Autocracy - - - - -0,09***

(-3,43)

-

Democracy - - - - - 0,07***

(2,37)

R2 0,66 0,71 0,68 0,68 0,66 0,67

Adjusted R2 0,65 0,69 0,67 0,67 0,65 0,66

N 128 73 121 127 121 121

Values in parentheses indicate t-statistics. ***, ** and * denote that parameter estimates are statistically significant at 1, 5% and 10% level.

On the other hand, such results can be explained by the limited data sample, which

doesn‟t allow us to account for all possible income levels. In our sample, countries which

predominantly lack data on the Gini coefficient are the low-income countries: While there are

25 observations on high-income countries and 40 on middle-income countries, there are only

8 on the low-income countries for the year 2006 (Appendix D). Such a limitation may hinder

us from observing the overall impact of income inequality on the environmental performance

due to the small number of observations on the starting point of the curve. It should be

mentioned that the average Gini coefficient is 41,72 for the low-income, 42,84 for the middle-

income and 31,55 for the high-income countries. 26

Heenrik et al (2001), p.365 27 Estimations of the original Kuznets curve yield:

; parameters are statistically significant at 1% level

23

Table 5.1.2 Testing for the low- and high-income country effects for the year 2006

Model specification Explanatory variable I II III IV

Constant

6,25*

(1,93)

12,9***

(4,59)

8,14**

(1,99)

12,09***

(4,38)

ln GDP per capita -1,16

(-0,98)

-3,47***

(-3,43)

-1,92

(-1,38)

-3,16***

(-3,18)

(ln GDP per capita)2

0,16

(1,10)

0,43***

(3,58)

0,26

(1,66)

0,39***

(3,31)

(ln GDP per capita)3 -0,006

(-1,07)

-0,02***

(-3,60)

-0,01*

(-1,83)

-0,02***

(-3,32)

Gini*low-income 0,004

(1,49)

- - -

Gini*middle-income 0,004**

(2,63)

- - -

Gini*high-income 0,004*

(1,84)

- - -

Polity*low-income - -0,01

(-0,79)

- -

Polity*middle-income - 0,009***

(3,85)

- -

Polity*high-income - 0,006***

(4,14)

- -

FreedomHouse*low-

income

- - 0,03

(1,14)

-

FreedomHouse*middle-

income

- - 0,02***

(2,69)

-

FreedomHouse*high-

income

- - 0,02***

(3,49)

-

Democracy*low-income - - - -0,02

(-0,19)

Democracy*middle-

income

- - - 0,09***

(2,74)

Democracy*high-income - - - 0,07**

(2,32)

R2 0,71 0,69 0,68 0,68

Adjusted R2 0,68 0,68 0,67 0,66

N 73 121 127 121

Values in parentheses indicate t-statistics. ***, ** and * denote that parameter estimates are statistically significant at 1, 5% and 10% level.

Following Boyce and Torras (1998), we also test for the possibility that income

inequality and the level of political freedom effects may differ depending on the country‟s

level of development. Since the level of development is associated with the level of income,

we create three dummy variables for each of the income group defined in the Data section.

We interact these dummies with the income inequality and political freedom variables; and

present the estimation results in Table 5.1.2. Unfortunately, we were unable to estimate the

impact of Autocracy for each of the income groups due to insufficient number of observations.

From these estimations we can conclude that the level of political freedom has the same

level of impact on the environmental performance in middle- and high-income countries,

24

since the coefficients of the interaction variables for these countries are statistically significant

and have similar magnitudes for Polity IV, Freedom House, and Democracy indicators. In

low-income countries we find no significant impact of neither income inequality nor the

measures capturing political freedom. This contradicts Boyce and Torras (1998) findings,

who conclude that political rights have a stronger effect on environmental degradation in low-

income countries. This can be partially explained by our limited sample that contains only 15

observations for low-income countries. Another explanation is the difference in the

methodology, since Boyce and Torras (1998) use a 5000 dollars per capita threshold to

distinguish between the low- and high-income countries. However, such results can also

suggest that on initial levels of economic development there is not much demand for the

environmental quality in the first place. While there may be some rich and political influential

interests in low-income countries, a predominant part of the population in low-income

countries can barely satisfy their basic needs. Even with the higher levels of political freedom,

there won‟t be any support for less environmental degradation until certain level of basic

needs‟ satiation is reached, i.e., until the majority of people have the ability to pay for the

environmental quality. Only then the ability to influence state policy becomes important in

determining the environmental performance of a country.

The same type of argument applies to the income inequality, which according to our

estimations has a strong positive impact on the environmental performance in middle- and

high-income countries, but not in the low-income countries; income effects become

insignificant. In low-income countries, the income groups that are rich enough to satisfy their

basic needs and to care about the environmental quality, typically are closely connected to the

business sector and to the politicians, and are more prone to favour profit-maximizing

incentives that lead to environmental degradation. The results of other studies are also mixed

in this respect: Boyce and Torras (1998) find contradicting impacts of income inequality in

low-income countries for different environmental indicators; Clement and Meunie (2010)

don‟t find any significant income inequality effects for SO2 emissions, but conclude that

increase in income inequality leads to more water pollution.

As a next step, we test for combined effects of income inequality and the level of political

freedom, including several additional explanatory variables at a time. The estimation results

are presented in Table 5.1.3. The Gini coefficient is included in each regression, and as in

previous estimations, all the income effects lose their statistical significance, while income

inequality effects are significant and positive. Once the Gini coefficient is taken into account,

we find no significant effects of the level of political freedom. Among all these measures,

25

only the Autocracy dummy has statistically significant effect, with a negative impact on the

environmental performance. Following Clement and Meunie (2010), we also estimate model

specifications that include interaction terms between income inequality and the

Democracy/Autocracy dummies. The results are presented in specification V and VI in Table

5.1.3 The interaction terms between inequality and dummies for democracy and autocracy are

insignificant, which suggests that the impact of income inequality does not depend on the

political regime. These results are in line with Clement and Meunie (2010), who find no

significant effect of the interaction terms when estimating the EKC for water pollutants.

Table 5.1.3 Combined effects of income inequality and political freedom on the environmental

performance for the year 2006

Model specification Explanatory variable I II III IV V VI

Constant

6,02*

(1,78)

6,28*

(1,73)

5,77*

(1,72)

6,26*

(1,80)

6,36*

(1,88)

5,97*

(1,70)

ln GDP per capita -1,08

(-0,86)

-1,16

(-0,87)

-1,00

(-0,81)

-1,16

(-0,90)

-1,23

(-0,98)

-1,11

(-0,86)

(ln GDP per capita)2

0,15

(0,96)

0,16

(0,97)

0,14

(0,95)

0,16

(1,00)

0,17

(1,12)

0,15

(0,97)

(ln GDP per capita)3 -0,006

(-0,95)

-0,01

(-0,96)

-0,01

(-0,98)

-0,01

(-0,97)

-0,007

(-1,15)

-0,006

(-0,96)

Gini coefficient 0,004**

(2,57)

0,004***

(2,65)

0,003**

(2,01)

0,005**

(2,63)

0,003*

(1,74)

0,007*

(1,82)

Polity 0,001

(0,32)

- - - - -

Freedom House - -0,004

(-0,27)

- - - -

Autocracy - - -0,10*

(-1,87)

- -1,03***

(-2,81)

-

Democracy - - - -0,04

(-0,74)

- 0,09

(0,47)

Gini*Autocracy - - - - 0,03

(2,74)

-

Gini*Democracy - - - - - -0,003

(-0,72)

R2 0,71 0,71 0,72 0,71 0,73 0,71

Adjusted R2 0,68 0,69 0,70 0,69 0,79 0,68

N 70 72 70 70 70 70

Values in parentheses indicate t-statistics. ***, ** and * denote that parameter estimates are statistically significant at 1, 5% and 10% level.

26

5.2 Estimations of the EKC hypothesis for the year 2008

The estimation results for the year 2008 are presented in Tables 5.2.1 and 5.2.2.

Table 5.2.1 presents the results for the following model specifications: Specification I is

our basic model, specification II includes the environmental effects of income inequality,

specification III includes the Polity IV index and specification IV uses the Freedom House

index. In specification V and VI we use the two dummy variables Autocracy and Democracy.

The EKC hypothesis is confirmed for all model specifications, except for the one

including the Gini coefficient. The explanatory power for the alternative model specifications

remains on the same high level. Both the Polity IV and Freedom House measures of the level

of political freedom are statistically significant and indicate that the environmental quality

improves with the increase in the level of political freedom. It should be also noted that while

the size of the per capita income effects changes substantially from the year 2006 to the year

2008, the effect of the level of political freedom remains of the fairly same size. The

Autocracy measure has the same negative sign as for the 2006 sample, but becomes

statistically significant for the 2008 estimations. This may be partially explained by the

increased number of observations in the sample. In general, we can conclude that the

modifications of the EPI methodology don‟t significantly influence the relationship between

the studied variables, compared to the year 2006.

We have 64 observations for the Gini coefficient, which contains only one low-income

country (Mozambique), 25 middle-income countries and 37 high-income countries. The Gini

coefficient is again on average 10 percentage points higher for the middle-income countries:

The average value for middle-income and high-income countries is 40,95, and 32,46,

respectively.

It may be the exclusion of all but one poor country and the small sample for middle-

income countries that drives statistically insignificant results when we estimate the inequality

effects on the environmental performance (Specification II in Table 5.2.1). Since almost all

observations for the low per capita income countries are excluded in our estimations, the EKC

relationship cannot be supported even if it prevails. The other explanation that we might be

implicitly estimating the original Kuznets curve, may also explain the poor support of the

EKC specification. Similarly to the year 2006, the original Kuznets curve for the year 2008

holds.28

28

Estimations of the original Kuznets curve yield: ; parameters are statistically significant at 1% level

27

Table 5.2.1 Estimations of the EKC hypothesis with additional variables for the year 2008

Model specification Explanatory variable I II III IV V VI

Constant

7,51***

(3,48)

-0,88

(-0,30)

7,18***

(3,12)

7,12***

(3,51)

7,90***

(3,24)

7,43***

(3,36)

ln GDP per capita -1,59**

(-2,04)

1,37

(1,35)

-1,51*

(-1,80)

-1,49**

(-2,04)

-1,76**

(-1,98)

-1,60*

(-1,96)

(ln GDP per capita)2

0,22**

(2,39)

-0,12

(-1,04)

0,22**

(2,16)

0,21**

(2,45)

0,24**

(2,30)

0,23**

(2,29)

(ln GDP per capita)3 -0,01**

(-2,57)

0,004

(0,84)

-0,01**

(-2,37)

-0,01***

(-2,71)

-0,01**

(-2,46)

-0,01**

(-2,46)

Gini coefficient - 0,001

(1,47)

- - - -

Polity - - 0,005**

(2,56)

- - -

Freedom House - - - 0,02***

(2,63)

- -

Autocracy - - - - -0,04

(-1,38)

-

Democracy - - - - - 0,05*

(1,89)

R2 0,64 0,73 0,66 0,66 0,64 0,65

Adjusted R2 0,63 0,71 0,65 0,65 0,63 0,64

N 144 63 138 144 138 138

Values in parentheses indicate t-statistics. ***, ** and * denote that parameter estimates are statistically significant at 1, 5% and 10% level.

Table 5.2.2 presents the estimation results of testing whether the effects of income

inequality and the level of political freedom differ depending on the country‟s level of

economic development. Similarly to the year 2006, we find that the level of political freedoms

has a positive and a statistically significant effect of similar magnitude in both middle- and

high-income countries: this is valid for the Polity IV index, the Freedom House index, and for

the Democracy measure. The same measures have a negative effect in the low-income

countries, which is statistically significant only for the Polity IV measure. Increased number

of observations on the Autocracy measure, compared to the year 2006, allows us to estimate

the impact of the autocratic regime on the environmental performance in countries with

different level of income. While this impact appears to be significant for the low- and high-

income countries, these results are hard to interpret, since most of the observations on the

countries with autocratic regime in our sample belong to the middle-income countries, while

there are only three for the high-income countries (Oman, Saudi Arabia and United Arab

Emirates) and only one for the low-income country (Eritrea). The estimations including the

Gini coefficient again don‟t provide any significant results. Moreover, since we only have one

observation of the Gini coefficient for low-income countries, the results for this country group

28

in model specifications including income inequality may not be representative for the low-

income country group and only capture country-specific effects in Mozambique.

When testing for the joint impact of income inequality and the level of political freedom

on the environmental performance, we don‟t obtain any statistically significant results for the

year 2008. Given our data sample, we couldn‟t identify neither income nor income inequality

or political rights effects. Therefore, we don‟t present our estimations here.

Table 5.2.2 Testing for the low- and high-income country effects for the year 2008

Model specification Explanatory variable I II III IV V

Constant

1,65

(0,42)

9,14***

(4,31)

8,81***

(4,07)

6,75***

(3,07)

8,95***

(3,33)

ln GDP per capita 0,55

(0,41)

-2,17***

(-2,78)

-2,02***

(-2,65)

-1,34*

(-1,69)

-2,08**

(-2,14)

(ln GDP per capita)2

-0,03

(-0,22)

0,29***

(3,09)

0,27***

(3,00)

0,19**

(2,06)

0,28**

(2,40)

(ln GDP per capita)3 0,0005

(0,09)

-0,01***

(-3,26)

-0,01***

(-3,20)

-0,008**

(-2,26)

-0,01**

(-2,54)

Gini*low-income -0,0005

(-0,48)

- - - -

Gini*middle-income 0,001

(1,47)

- - - -

Gini*high-income 0,001

(1,53)

- - - -

Polity*low-income - -0,02**

(-2,45)

- - -

Polity*middle-income - 0,005**

(2,49)

- - -

Polity*high-income - 0,007***

(4,29)

- - -

Freedom House*low-income - - -0,003

(-0,17)

- -

Freedom House*middle-

income

- - 0,02*

(2,68)

- -

FreedomHouse*high-income - - 0,02**

(2,51)

- -

Autocracy*low-income - - - 0,16***

(4,22)

-

Autocracy*middle-income - - - -0,02

(-0,72)

-

Autocracy*high-income - - - -0,19***

(-6,24)

-

Democracy*low-income - - - - -0,09

(-1,31)

Democracy*middle-income - - - - 0,07**

(2,38)

Democracy*high-income - - - - 0,08**

(2,36)

R2 0,73 0,68 0,66 0,66 0,67

Adjusted R2 0,70 0,67 0,65 0,65 0,66

N 63 138 144 138 138

Values in parentheses indicate t-statistics. ***, ** and * denote that parameter estimates are statistically significant at 1, 5% and 10% level.

29

Conclusions

The purpose of this essay was to assess the EKC hypothesis using the Environmental

Performance Indices, as well as to investigate whether income inequalities and the difference

in the level of political freedoms influence the EKC shape. We performed OLS estimations on

cross-section data for the 2006 and 2008 EPI indices. We found strong support for the EKC

hypothesis, which means that environmental performance initially worsens but eventually

improves with an increasing level of per capita income. Moreover, our estimations confirm

that higher level of political freedom improves environmental performance. However, this

effect comes into power only once a certain income level is reached: we find no significant

impact of political freedom on the environmental performance in the countries with the lowest

income levels, but only for the middle- and higher-income countries.

Our study contributes to the EKC literature by using not a separate, but a composite

indicator of environmental quality, which EPI represents. In contrast to the only study

deploying the EPI by Yoshioka (2010), we find strong support of the EKC hypothesis, which

can be explained by a better model specification.

The absence of time-series data for the Environmental Performance Index doesn‟t allow

us to track the changes in the environmental performance over time. However, the data

available still gives us a “snapshot” picture of the relationship between the environmental

performance and per capita income around the world, and allows us to investigate some

aspects of the EKC that have not, to our knowledge, been previously investigated.

Our estimation results suggests that, it is only after the initial stages of economic

development and strong economic progress in terms of economic growth has already taken

place, that democratization can help remedy environmental problems. Only at this stage,

increased political rights and liberties, which provide better means for people to influence the

environmental quality, are able to mitigate the negative environmental impacts of economic

growth. This contradicts Boyce and Torras (1998) findings, who conclude that political rights

have a stronger effect on environmental degradation in low-income countries, but may reflect

that the authors of the mentioned study included both low-income and low middle-income

countries in their definition of low-income countries. Thus, our findings may be more

accurate in this respect.

We also conclude that including the measure of income inequality in the EKC model may

distort the results of estimations, since such a specification may implicitly estimate the

30

original Kuznets curve, i.e. the impact of income on income inequality, and hence on the

environmental degradation. These findings are in line with Heenrik et al (2001).

The future access to better data availability will allow the further study of how the EKC

develops over time and how income inequality affects its shape at all levels of economic

development.

31

References

Published sources:

Bimonte, S. (2002). Information access, income distribution, and the Environmental Kuznets Curve,

Ecological Economics, 41:145–156.

Borghesi, S., and Vercelli, A. (2003). Sustainable globalisation, Ecological Economics, 44:77–89.

Borissov, K., Brechet, T., and Lambrecht, S. (2010). Median voter environmental maintenance, CORE

Discussion Paper.

http://www.uclouvain.be/cps/ucl/doc/core/documents/E2M2_108.pdf (April 2011)

Boyce, J. K. (1994). Inequality as a cause of environmental degradation, Ecological Economics,

11:169–178.

Boyce, J.K., and Torras, M. (1998). Income, Inequality, and Pollution: A Reassessment of the

Environmental Kuznets Curve, Ecological Economics, 25:147–160.

Buitenzorgy, M., and Mol, A.P.J. (2011). Does Democracy Lead to a Better Environment?

Deforestation and the Democratic Transition Peak, Environmental and Resource Economics,

48(1):59-70.

Chimeli, A., Braden, J. (2009). A capital scarcity theory of the environmental Kuznets curve,

Environment and Development Economics, 14: 541–564.

Clement, M., and Meunie, A. (2010). Is Inequality Harmful for the Environment? An Empirical

Analysis Applied to Developing and Transition Countries, Review of Social Economy, 68 (4):

413-445.

Coondoo, D. and Dinda, S. (2008). The carbon dioxide emission and income: a temporal analysis of

cross-country distributional patterns, Ecological Economics, 65 (2008), pp. 375–385.

Dasgupta, S., and Bnoit, L. (2002). Confronting the Environmental Kuznets Curve, Journal of

Economic Perspective, 16(1):147-168.

Dinda, S. (2004). Environmental Kuznets Curve Hypothesis: A Survey, Ecological Economics,

49:431–455.

Emerson, J., D. C. Esty, M.A. Levy, C.H. Kim, V. Mara, A. de Sherbinin, and T. Srebotnjak. (2010).

2010 Environmental Performance Index. New Haven: Yale Center for Environmental Law and

Policy.

Esty, D. C., Levy, M., Srebotnjak, T., de Sherbinin, A., Kim, C., Anderson, B. (2006).

Pilot 2006 Environmental Performance Index, Yale Center for Environmental Law and Policy

and Center for International Earth Science Information Network (CIESIN), Columbia

University.

Esty, D. C., Levy, M.A., Kim, C.H., de Sherbinin, A., Srebotnjak, T., and Mara, V. (2008).

2008 Environmental Performance Index. New Haven: Yale Center for Environmental Law and

Policy.

Fare, R., Grosskopf, S., Hernandez-Sancho, F. (2004). Environmental performance: an index number

approach, Resource and Energy Economics, 26: 343-352.

Gallagher, K.P., and Thacker, S.C. (2008). Democracy, income, and environmental quality, Working

paper 164, Political Economy Research Institute, University of Massachusetts.

Grossman, G.M., and Krueger, A.B., (1993). Environmental impacts of the North American Free

Trade Agreement. In: Garber, P. (Ed.), The U.S. –Mexico Free Trade Agreement. MIT Press,

Cambridge, pp. 13–56.

Grossman, G.M., and Krueger, A.B., (1995). Economic growth and the environment, The Quarterly

Journal of Economics, 110:353–377.

Gurr, T.D., Jaggers, K., Marshall, M. G., (2010). Polity IV Project: Political Regime Characteristics

and Transitions, 1800-2009. Dataset Users‟ Manual, Center for Systemic Peace.

http://www.systemicpeace.org/inscr/p4manualv2009.pdf (May 2011)

32

He, J. (2007). Is the environmental Kuznets curve hypothesis valid for developing countries? A survey,

Working Paper 07–03, University de Sherbrooke.

Heerink, N., Mulatu, A. and Bulte, E. (2001). Income inequality and the environment: aggregation

bias in environmental Kuznet curves, Ecological Economics, 38: 359–367.

Hill, R. J., and Magnani, E. (2002). An exploration of the conceptual and empirical basis of the

Environmental Kuznets Curve, Australian Economic Papers, 41(2): 239–254.

Holland, T. G., Peterson, G. D., and Gonzalez, A. (2009). A cross-national analysis of how economic

inequality predicts biodiversity loss, Conservation Biology, 23: 1304–1313.

Jha, R., Bhanu Murth, K.V., (2003). An inverse global environmental Kuznets curve. Journal of

Comparative Economics, 31: 352–368.

Kempf, H., and Rossignol, S. (2007). Is inequality harmful for the environment in a growing

economy? Politics and Economics, 19(1): 53-71.

Kijima, M., Nishide, K., Ohyama A. (2010). Economic models for the environmental Kuznets curve: a

survey. Journal of Economic Dynamics & Control, 34:1187–1201.

Koop, G., and Tole, L. (2001). Deforestation, distribution and development, Global Environmental

Change, 11:193–202.

Kuznets, S. (1955). Economic growth and income inequality, American Economic Review, 45: 1-28.

Li, Q., and Reuveny, R. (2006). Democracy and Environmental Degradation, International Studies

Quarterly, 50(4): 935–956.

Magnani, E. (2000). The Environmental Kuznets Curve, environmental policy and income distribution,

Ecological Economics, 32: 431–443.

Magnani, E., (2001). The environmental Kuznets curve: Development path or policy result?

Environmental Modelling and Software, 16:157–166.

Marsiliani, L., and Renström, T. (2002), On Income Inequality and Green Preferences, Discussion

paper CEPR 3677, and W. Allen Wallis Institute of Political Economy, University of Rochester,

Working Paper no. 30/2002.

McConnell, K. (1997). Income and demand for environmental quality, Environment and development

Economics, 2:383-400.

Moomaw, W.R., Unruh, G.C., (1997). Are Environmental Kuznets curves misleading us? The case of

CO2 emissions, Environment and Development Economics, 2: 451– 463.

Panayotou, T., (1993). Empirical tests and policy analysis of environmental degradation at different

stages of economic development, ILO, Technology and Employment Programme, Geneva.

Panayotou, T., (1997). Demystifying the environmental Kuznets curve: turning a black box into a

policy tool, Environment and Development Economics 2: 465– 484.

Panayotou, T. (1999). The economics of environments in transition, Environment and Development

Economics, 4(4):401– 412.

Panayotou, T. (2003). Economic Growth and the Environment, Chapter 2, Economic Survey of

Europe, No.2:45-67.

http://www.unece.org/ead/pub/032/032_c2.pdf (Accessed on April 21, 2011)

Raymond, L (2004). Economic growth as environmental policy? Reconsidering the Environmental

Kuznets Curve, Journal of Public Policy, 24: 27–48.

Shafik, N., Bandyopadhyay, S., (1992). Economic Growth and Environmental Quality: Time Series

and Cross-Country Evidence. Background Paper for the World Development Report. The World

Bank, Washington, DC.

Stern, D.I, (2003). The Environmental Kuznets Curve, International Society for Ecological Economics,

Internet Encyclopedia of Ecological Economics, Department of Economics, Rensselaer

Polytechnic Institute, Troy, NY 12180.

33

Stern, D.I. (2004). The rise and fall of the environmental Kuznets curve, World Development, 32(8):

1419-39.

Verbeek, M. (2004). A Guide to Modern Econometrics (2nd

ed.). John Wiley & Sons, Ltd.

Vukina, T., Beghin, J.C., Solakoglu, E.G., (1999). Transition to markets and the environment: effects

of the change in the composition of manufacturing output, Environment and Development

Economics, 4(4):582– 598.

Xiaoyu, L., Jun, Y., Pengfei, S. (2011). Structure and Application of a New Comprehensive

Environmental Pollution Index, Energy Procedia, 5: 149-154.

Yoshioka, S. (2010). Estimation of Environmental Kuznets Curve for Various Indicators: Evidence

from Cross-Section Data Analysis, Tounan Asia Kenkyuu Nenpou, 51:1-18.

http://naosite.lb.nagasaki-u.ac.jp/dspace/bitstream/10069/23345/1/toasia51_1.pdf (2011-04-21)

Zaim, O., F. Taskin, (2000). A Kuznets curve in environmental efficiency: an application on OECD

countries, Environmental and Resource Economics, 17: 21-36.

Electronic sources:

http://data.worldbank.org/ - World Development Indicators database from World Bank

http://www.systemicpeace.org/polity/polity4.htm - Polity IV Project of The Center for Systemic Peace

http://freedomhouse.org/ – Freedom House

http://epp.eurostat.ec.europa.eu - Eurostat - the statistical office of the European Union

http://www.oecd.org/ - OECD statistics

34

Appendix A EPI framework

Table A1. EPI 2006 Framework

Overal Performance

Broad Objectives

Policy Categories Indicators

EPI Environmental Health

Environmental Health

Child Mortality

Indoor air pollution

Outdoor air pollution

Drinking water

Adequate sanitation

Urban particulates

Ecosystem Vitality

Air quality

Regional Ozone

Water resources Nitrogen loading

Water Consumption

Biodiversity & Habitat

Wilderness Protection

Ecoregion Protection

Timber Harvest Rate

Productive natural resources

Agricultural subsidies

Overfishing

Sustainble energy Energy efficiency

Renewable energy

CO2 per GDP

Source: EPI 2006

35

Table A2. EPI 2008 Framework

Index ObjectivesObjective

Weight (%

of EPI)

Policy Categories

Policy

Category

Weight

(% of EPI)

Subcategories

Sub-

category

Weight (%

of EPI)

IndicatorsIndicator

Weight in

EPI %

EPIEnvironmental

Health50 Environmental Health 50

Environmental burden

of disease25

Environmental burden of

disease (DALYs)25

Water (effects on

humans)12,5 Adequate sanitation 6,25

Drinking water 6,25

Air Pollution 12,5 Urban particulates 5

(effects on humans) Indoor air pollution 5

Local ozone 2,5

Ecosystem

Vitality50

Air Pollution (effects

on nature)2,5

Air Pollution (effects

on nature)2,5 Regional ozone 1,25

Sulfur dioxide emissions 1,25

Water (effects on

nature)7,5

Water (effects on

nature)7,5 Water quality 3,75

Water stress 3,75

Biodiversity & Habitat 7,5 Biodiversity & Habitat 7,5 Conservation risk index

Effective conservation

Critical habitat protection

Marine Protected Areas

Productive Natural

Resources7,5 Forestry 2,5 Growing stock change 2,5

Fisheries 2,5 Marine Trophic Index 1,25

Trawling intensity 1,25

Agriculture 2,5 Irrigation Stress 0,5

Agricultural Subsidies 0,5

Intensive cropland 0,5

Burnt Land Area 0,5

Pesticide Regulation 0,5

Climate Change 25 Climate Change 25 Emissions per capita 8,333

Emissions per electricity

generation8,333

Industrial carbon intensity 8,333

7,5

Source: EPI 2008 http://epi.yale.edu:2008/Home

36

Appendix B Descriptive statistics

Table B1. EPI 2006 Descriptive statistics

Mean Median Maximum Minimum Std. Dev. Obs

GDP per capita 11161,81 5781,00 51449,17 271,55 12514,53 128

log(GDP per capita) 8,59 8,66 10,85 5,60 1,31 128

GINI coefficient (extended) 38,75 37,72 58,49 23,7 9,45 73

Freedom House 4,58 5,00 7,00 1,00 1,91 132

Polity IV 3,84 7,00 10,00 -10,00 9,58 127

EPI 64,45 64,60 88,00 25,70 6,38 126

Environmental Health 63,39 69,20 99,40 0,00 29,40 133

Child Mortality 73,05 91,80 99,60 0,00 32,93 133

Indoor Air Pollution 52,51 52,00 100,00 0,00 40,14 133

Drinking Water 66,32 74,70 100,00 0,00 32,02 133

Adequate Sanitation 60,94 61,10 100,00 0,00 33,84 133

Air Quality 54,38 55,50 98,00 6,90 18,96 133

Urban Particulates 64,77 71,90 96,20 0,00 24,80 133

Regional Ozone 43,98 31,60 100,00 0,00 28,13 133

Water Resources 81,91 91,70 100,00 6,50 21,77 133

Nitrogen Loading 89,69 98,60 100,00 0,00 24,14 133

Water Consumption 74,13 84,00 100,00 0,00 30,40 133

Biodiversity and Habitat 50,70 50,90 88,10 5,10 18,95 133

Wilderness Protection 19,71 14,40 72,50 0,00 17,71 133

Ecoregion Protection 62,85 69,90 100,00 0,00 31,45 133 Productive Resource Management 74,08 77,30 100,00 33,30 19,08 133

Timber Harvest Rate 89,84 100,00 100,00 0,00 25,62 133

Agricultural Subsidies 82,92 100,00 100,00 0,00 32,64 133

Overfishing 37,75 33,30 83,30 0,00 22,11 102

Sustainable Energy 66,52 74,80 92,40 0,00 22,77 133

Energy Efficiency 72,51 80,20 100,00 0,00 27,09 133

Renewable Energy 16,81 9,20 100,00 0,00 20,98 133

C02 per GDP 71,96 82,50 98,10 0,00 26,09 133

37

Table B2. EPI 2008 Descriptive statistics

Mean Median Max Min Std. Dev. Obs

GDP per capita 11975,51 6951,24 73126,70 289,94 13287,00 144

log(GDP per capita) 8,70 8,85 11,20 5,67 1,29 144

GINI coefficient (extended) 36,04 34,21 54,46 23,40 8,74 63

Freedom House 4,63 4,75 7,00 1,00 1,90 148

Polity IV 4,04 7,00 10,00 -10,00 6,30 143

EPI 71,87 74,10 95,51 39,05 12,74 149 1. Environmental Health 74,56 84,51 99,38 6,02 24,78 149 Environmental Health 74,56 84,51 99,38 6,02 24,78 149

1.1 Environmental burden of disease (DALYs)

79,86 92,79 99,82 0,00 27,24 149

1.1.1 Environmental burden of disease (DALYs)

79,86 92,79 99,82 0,00 27,24 149

1.2 Water (effects on humans) 67,96 75,72 100,00 0,00 29,95 149

1.2.1 Adequate sanitation 64,76 73,10 100,00 0,00 32,32 149

1.2.2 Drinking water 71,16 83,02 100,00 0,00 30,25 149

1.3 Air Pollution (effects on humans) 70,58 72,49 97,89 15,99 23,23 149

1.3.1 Urban particulates 73,25 83,76 100,00 0,00 28,23 149

1.3.2 Indoor air pollution 57,31 68,42 100,00 0,00 37,95 149

1.3.3 Local ozone 91,76 99,78 100,00 0,00 21,62 149

2. Ecosystem Vitality 69,18 71,30 92,81 37,09 10,76 149 Air Pollution (effects on nature) 90,72 96,14 99,97 43,99 13,36 149

2.1 Air Pollution (effects on nature) 90,72 96,14 99,97 43,99 13,36 149

2.1.1 Regional ozone 90,26 99,86 100,00 0,00 22,87 149

2.1.2 Sulfur dioxide emissions 91,18 96,73 99,94 0,00 16,64 149

2.2 Water (effects on nature) 66,58 67,64 98,98 0,00 18,49 149

2.2.1 Water quality 48,86 49,55 99,02 0,00 26,14 149

2.2.2 Water stress 84,29 93,80 100,00 0,00 20,43 149

Biodiversity & Habitat 46,36 46,48 100,00 0,23 27,56 149

2.3. Biodiversity & Habitat 46,36 46,48 100,00 0,23 27,56 149

2.3.1 Conservation risk index 63,15 72,65 100,00 0,00 34,16 149

2.3.2 Effective conservation 45,77 42,11 100,00 0,00 34,51 149

2.3.3 Critical habitat protection 45,73 45,65 100,00 0,00 32,00 65

2.3.4 Marine Protected Areas 33,80 9,00 100,00 0,00 42,45 149

Productive Natural Resources 79,82 82,29 98,97 44,39 10,86 149

2.4. Forestry 88,24 100,00 100,00 0,00 20,83 148

2.4.1 Growing stock change 88,24 100,00 100,00 0,00 20,83 148

2.5 Fisheries 70,29 77,23 99,53 0,00 22,61 113

2.5.1 Marine Trophic Index 84,33 98,46 100,00 0,00 23,71 99

2.5.2 Trawling intensity 59,87 72,31 99,06 0,00 31,93 113

2.6 Agriculture 77,72 77,98 99,14 46,49 10,55 140

2.6.1 Irrigation Stress 88,56 99,88 100,00 0,00 21,01 142

2.6.2 Agricultural Subsidies 79,14 100,00 100,00 0,00 33,37 149

2.6.3 Intensive cropland 79,20 90,18 100,00 0,00 26,35 145

2.6.4 Burnt Land Area 82,85 93,03 99,98 0,00 23,15 143

2.6.5 Pesticide Regulation 59,95 77,27 100,00 0,00 38,93 149

Climate Change 71,46 72,90 99,78 16,14 15,89 149

2.7 Climate Change 71,46 72,90 99,78 16,14 15,89 149

2.7.1 GHG Emissions per capita 83,34 89,44 100,00 0,00 20,83 149

2.7.2 Emissions per electricity generation

52,25 50,52 100,00 0,00 27,48 149

2.7.3 Industrial carbon intensity 78,79 84,96 100,00 0,00 23,42 149

38

Appendix C Groups of countries according to their per capita income

Table C1 Groups of countries according to their per capita income for the year 2006

Low-income Middle-income High-income

Burundi Central African Republic Dem. Rep. Congo Ethiopia Guinea Guinea-Bissau Madagascar Malawi Mozambique Nepal Niger Rwanda Sierra Leone Togo Uganda

Albania Algeria Angola Argentina Armenia Azerbaijan Bangladesh Benin Bolivia Brazil Bulgaria Burkina Faso Cambodia Cameroon Chad China Colombia Congo Costa Rica Côte d'Ivoire Dominican Rep. Ecuador Egypt El Salvador Gambia Georgia Ghana Guatemala Haiti Honduras India Indonesia Iran Jamaica Jordan Kazakhstan Kenya Kyrgyzstan

Laos Lebanon Mali Mauritania Moldova Mongolia Morocco Namibia Nicaragua Nigeria Pakistan Panama Papua New Guinea Paraguay Peru Philippines Romania Senegal South Africa Sri Lanka Sudan Suriname Swaziland Syria Tajikistan Tanzania Thailand Tunisia Turkey Turkmenistan Ukraine Uzbekistan Venezuela Viet Nam Yemen Zambia

Australia Austria Belgium Canada Chile Cyprus Czech Rep. Denmark Finland France Gabon Germany Greece Hungary Iceland Ireland Israel Italy Japan Malaysia Mexico Netherlands New Zealand Norway Oman Poland Portugal Russia Saudi Arabia Slovakia Slovenia South Korea Spain Sweden Switzerland Trinidad & Tobago United Arab Emirates United Kingdom United States

39

Table C1 Groups of countries according to their per capita income for the year 2008

Low-income Middle-income High-income

Burundi Central African Republic Dem. Rep. Congo Eritrea Ethiopia Guinea Guinea-Bissau Madagascar Malawi Mozambique Niger Sierra Leone Togo

Albania Algeria Angola Armenia Azerbaijan Bangladesh Belarus Belize Benin Bolivia Bosnia and Herzegovina Brazil Bulgaria Burkina Faso Cambodia Cameroon Chad China Colombia Congo Costa Rica Côte d'Ivoire Djibouti Dominican Rep. Ecuador Egypt El Salvador Fiji Georgia Ghana Guatemala Guyana Haiti Honduras India Indonesia Iran Iraq Jamaica Jordan Kazakhstan Kenya Kyrgyzstan

Laos Lebanon Macedonia Mali Mauritania Mauritius Moldova Mongolia Morocco Namibia Nepal Nicaragua Nigeria Pakistan Panama Papua New Guinea Paraguay Peru Philippines Romania Rwanda Senegal Solomon Islands South Africa Sri Lanka Sudan Swaziland Syria Tajikistan Tanzania Thailand Tunisia Turkey Turkmenistan Uganda Ukraine Uruguay Uzbekistan Venezuela Viet Nam Yemen Zambia

Argentina Australia Austria Belgium Botswana Canada Chile Croatia Cyprus Czech Rep. Denmark Estonia Finland France Gabon Germany Greece Hungary Iceland Ireland Israel Italy Japan Latvia Lithuania Luxembourg Malaysia Mexico Netherlands New Zealand Norway Oman Poland Portugal Russia Saudi Arabia Slovakia Slovenia South Korea Spain Sweden Switzerland Trinidad & Tobago United Arab Emirates United Kingdom United States

40

Appendix D Countries with the Gini coefficient available

Table D2. Year 2006

Low-income Middle-income High-income

Burundi Dem. Rep. Congo Guinea Mozambique Niger Rwanda Togo Uganda

Albania Argentina Armenia Azerbaijan Bolivia Brazil Bulgaria Cambodia Colombia Costa Rica Côte d'Ivoire Dominican Rep. Ecuador El Salvador Georgia Ghana Guatemala Honduras Indonesia Jordan

Kazakhstan Kyrgyzstan Laos Mali Moldova Mongolia Morocco Pakistan Panama Paraguay Peru Philippines Romania Sri Lanka Tanzania Thailand Turkey Ukraine Venezuela Viet Nam

Austria Belgium Chile Cyprus Czech Rep. Denmark Finland France Germany Greece Hungary Ireland Italy

Malaysia Mexico Netherlands Norway Poland Portugal Russia Slovakia Slovenia Spain Sweden United Kingdom

Table D2. Year 2008

Low-income Middle-income High-income

Mozambique Albania Armenia Azerbaijan Belarus Brazil Bulgaria Costa Rica Côte d'Ivoire Ecuador Georgia Indonesia

Laos Macedonia Moldova Mongolia Panama Paraguay Peru Romania Thailand Turkey Uganda Ukraine Uruguay Viet Nam

Argentina Australia Austria Belgium Chile Croatia Cyprus Czech Rep. Denmark Estonia Finland France Germany Greece Hungary Iceland Ireland Israel Italy

Latvia Lithuania Luxembourg Malaysia Mexico Netherlands New Zealand Norway Poland Portugal Russia Slovakia Slovenia South Korea Spain Sweden United States United Kingdom